Decoding Multi-Omics Signatures in Lower-Grade Glioma Using Protein-Protein Interaction-Informed Graph Attention...

Area of Science:

• Computational biology and bioinformatics
• Oncology and neuro-oncology
• Genomics and molecular biology

Background:

• Lower-grade gliomas (LGGs) exhibit significant biological and clinical heterogeneity.
• Molecular stratification is crucial for LGG diagnosis, prognosis, and treatment decisions.
• Unimodal classifiers fail to capture complex cross-layer regulatory dynamics in glioma.

Purpose of the Study:

• To develop a protein-protein interaction (PPI)-informed hybrid model for multi-omics data integration.
• To enhance molecular stratification and identify key biomarkers for lower-grade gliomas.
• To improve diagnostic and therapeutic strategies through explainable AI.

Main Methods:

• A hybrid model combining RNA expression, DNA methylation, and microRNA expression data.
• Integration of Graph Attention Network (GAT), Random Forest (RF), and logistic stacking ensemble learning.
• ElasticNet for feature selection, SMOTE for class imbalance, and cross-validation for performance assessment.

Main Results:

• The multi-omics model achieved superior subtype classification rates (up to 0.984 ± 0.012) compared to single-omics approaches.
• DNA methylation emerged as the most discriminative molecular data modality.
• Key subtype-specific biomarkers (UBA2, LRRC41, ANKRD53, WDR77) were identified with significant biological relevance.

Conclusions:

• The proposed multi-omics framework offers a robust computational approach for molecular stratification in LGG.
• The model provides explainable insights, facilitating biomarker discovery for clinical applications.
• This approach bridges predictive accuracy with biological understanding for improved patient outcomes.